34th INTERNATIONAL CONFERENCE ON
PRODUCTION ENGINEERING
29. - 30. September 2011, Niš, Serbia
University of Niš, Faculty of Mechanical Engineering
WIRELESS SENSOR NETWORK APPLICATION IN MONITORING OF MACHINING
OPERATIONS
Zivana Jakovljevica, Ph.D., Assistant Professor, Miroslav Pajicb, Ph.D. Candidate, Dragan Aleksendricc, Ph.D.,
Assistant Professor, Dragan Milkovicc, M.Sc., Assistant
a
Department of Production Engineering, Faculty of Mechanical Engineering, University of Belgrade, Kraljice Marije
16, 11 000 Belgrade, Serbia, bDepartment of Electrical and Systems Engineering, University of Pennsylvania, PA
19104, USA, cFaculty of Mechanical Engineering, University of Belgrade, Kraljice Marije 16, 11 000 Belgrade, Serbia
zjakovljevic@mas.bg.ac.rs, pajic@seas.upenn.edu, daleksendric@mas.bg.ac.rs, dmilkovic@mas.bg.ac.rs
Abstract: The scope of this paper is the research of the possibilities of application of Wireless Sensor
Networks (WSNs) for monitoring of machining operations. Due to the limited battery lifetime, WSNs are
energy constrained networks and energy efficiency of the network is a very important issue. Since
communication requires significantly more energy then processing, instead of sending raw data through
the network, data processing and decision making should be performed in the node. Consequently, as a
basis for WSN design we consider IEEE 802.15.4 Wireless Networking Standard, created for low data
rates and very low power consumption which makes it suitable for remote monitoring and control. In this
paper, a prototype of the node, designed for acceleration monitoring, has been developed. The node has
been created at the Faculty of Mechanical Engineering in Belgrade and it is based on low power Atmel
Atmega16 microcontroller and IEEE 802.15.4 compatible transceiver.
Key words: Wireless sensor networks, machining operations monitoring
1. INTRODUCTION
In-situ sensor monitoring of machining operations has
attracted significant attention in the last decade. During
machining, cutting forces, vibrations, acoustic emission,
temperature, audible sound, etc. are influenced by cutting
tool state and conditions of the material removal process.
The aforementioned process quantities can be measured
and acquired signals can be processed to extract features
correlated with the process and tool condition. These
features carry information for on-line decision making
and diagnosis of the cutting process state. After the
diagnosis, the appropriate actions are executed by human
operator or control system. A number of techniques for
tool condition and process parameters monitoring using
the described procedure have been proposed [1].
Nevertheless, the industrial application of on-line
condition-based monitoring systems is still sporadic.
The most significant limiting factors for implementation
of the proposed techniques are sensor deployment and
wiring. It is necessary to place the sensor very close to the
cutting zone; on the tool would be the best. Nevertheless,
the tools are placed in spindles or on moving platforms,
which makes them difficult or impossible to access by
wiring. Besides, when wiring is feasible, its cost is
estimated to 10-100$ per sensor [2].
Recent developments in wireless technology have opened
up new possibilities. Communication between the sensor
and the supervisory or control system can be wireless,
which enables sensor placement at very inaccessible
locations, such as tool (Fig. 1). In this scenario, the sensor
operates as a node in a Wireless Sensor Network (WSN).
It has embedded processor and radio transceiver. The
node is battery charged and energy efficient operation of
the sensor is of critical importance.
Fig. 1. Mounting of the wireless sensor to the tool
It is worth noting that radio communication (including
both transmissions and receptions) of a sensor node is
significantly more energy consuming than in-node
computation. To exploit benefits of wireless
communication, the sensor should be made intelligent. It
should carry out not only the measurements, but also data
acquisition, signal processing, features generation and
decision making. Instead performing a batch wireless
transmission of the raw signal and the processing at
central computer, the whole diagnosis process should be
performed at the node. In this case, only the information
about the decision (diagnosis) should be transmitted.
Machining operations monitoring assumes high sampling
rates. Sensor integrated tooling for monitoring of tool
vibration and temperature in milling using wireless
communication was studied in [3]. Nevertheless, contrary
to the above given observation, instead of in-node signal
processing, the raw data (20KHz) were sent to PC using
intensive Bluetooth (IEEE 802.15.1) communication. In
[4], the machine tool vibrations were monitored using
wireless accelerometer. The less communication intensive
IEEE 802.15.4 wireless standard was used for raw signal
transmission to the PC. Signal processing and decision
making were carried out in PC. Due to the bandwidth
constraint, the sampling rate was restricted to only 1000
Hz. In both applications, the communication is very
intensive which has repercussions on battery lifetime.
The utilization of wireless sensors assumes the in-node
signal processing using low-power microcontrollers.
There have been some approaches [5, 6] that used
microcontroller based systems for tool breakage detection
in end milling. Instead of transmitting a large amount of
data, the signal is analyzed at the node. On abnormal
event (tool breakage) detection, the message is sent to the
PC using SMS [5] or wired communication (Ethernet) [6].
Bluetooth (IEEE 802.15.1) and IrDA are designed for
high data rate applications such as voice, video and LAN
communication. They are very communication intensive
and thus suffer from high energy consumption, making
them unsuitable for wireless sensor applications such as
one at hand. GSM and GPRS are even less appropriate.
The energy consumption is also very high and there are
additional costs related to the GSM or GPRS modules
(~100$) and SIM card purchase and operation. On the
other hand, IEEE 802.15.4 standard offers low data rate
and low power. It is characterized by long battery life. For
these reasons this standard is used as a basis for WSN.
In this paper, the possibilities to employ WSNs in
machining operations monitoring have been investigated.
An overview of WSN was presented as well as a
prototype node developed at Faculty of Mechanical
Engineering, University of Belgrade (FME). The node is
employed in tool acceleration measurements in turning. In
this example, the decision making is carried out at the
node and only the information about significant event is
transmitted to the coordinator.
2. INTRODUCTION TO WSNs
In a Wireless Sensor Network (WSN) each node has one
or more sensors, embedded processors and low-power
radio. All the nodes coordinate in order to perform a
common task. In-network data processing greatly reduces
the energy consumption. It can be additionally reduced by
applying appropriate communication protocols. IEEE
802.15.4 can operate at star or peer to peer topology.
IEEE 802.15.4 has three operating frequency bands with
different number of channels and transfer rates: 868 MHz
 1 channel – 0 (20Kb/s); 915 MHz  10 channels – 110 (40Kb/s); 2.4 GHz  16 channels – 11-26 (250Kb/s).
In the IEEE 802.15.4 architecture three layers can be
distinguished (Fig. 2). These are: 1) Physical (PHY) layer,
2) Medium Access Control (MAC) layer and 3) Network
and Application layer.
Fig. 2. IEEE 802.15.4 architecture
Physical layer has the following functionalities [7]: 1)
activation and deactivation of the radio transceiver, 2)
energy detection within the current channel, 3) link
quality indication for received packets, 4) clear channel
assessment (CCA), 5) channel frequency selection, 6)
data transmission and reception. PHY frame structure of
transmitted data is shown in Fig. 3.
Fig. 3. PHY frame structure [7]
In IEEE 802.15.4 MAC layer two modes of operation are
defined: 1) beacon enabled network, 2) non-beacon
enabled network. In the beacon mode (Fig. 4a) the device
waits for the coordinator’s beacon to start transmission.
When the transmission is complete, the coordinator
schedules the next beacon and the device goes to sleep.
All the devices in the network know when to
communicate with each other. The timing circuits of the
nodes have to be synchronized. In the non-beacon mode
all nodes, except the coordinator, are almost always
‘asleep’. The device wakes up on event and sends a
message to the coordinator. Beacon mode is used when
sink runs on batteries, while non-beacon is preferred
when the sink is non energy constrained (in this mode the
coordinator has to always ‘listen’).
IEEE 802.15.4 MAC layer enables the creation of
different MAC protocols that define the communication
procedure between nodes. The MAC protocols must be
energy efficient. There are many sources of energy waste
in WSN, and the most important are [8]: 1) Collided
packets: when more than one packet is sent to a node at
the same time; since the receiver can not receive both
packets, these packets are rejected and should be
retransmitted; 2) Overhearing: node receives packets
intended to other nodes; 3) Control packet overhead:
minimal control packets should be used in transmission;
4) Idle listening: listening of the idle channel to receive
possible traffic; 5) Overemitting: sending the message
when the destination node is not ready.
There are many different MAC protocols [8]: B-MAC,
Sensor-MAC, WiseMAC, Traffic Adaptive MAC
protocol, D-MAC, etc. Nevertheless, no MAC protocol is
accepted as a standard and it is application dependant.
In monitoring operations, communication is usually
triggered by a sensing event. Therefore, if no sensing
event occurs the nodes could stay idle for a long time. To
preserve energy, radio transceivers are programmed to
stay in the sleep mode, thus avoiding idle listening. This
can be easily implemented in star topologies, where all
nodes communicate with the coordinator (i.e., gateway) in
a single hop. Nevertheless, the reach and the presence of
obstacles in the node’s vicinity usually demand the multihop between the sensor node and the sink. In this case, the
objective is to reliably deliver data to the sink (i.e.,
gateway), with minimum latency, while consuming the
minimal amount of energy.
B-MAC protocol [9], used in this paper, is designed for
monitoring operations. It provides nodes with duty cycle
mechanism in order to eliminate idle listening which is
very important in multi-hop networks. It is convenient for
monitoring of machining operations as well as for
machine tools control.
Radio duty cycling [9] is very important for the reliable
data transmission with low latency. In B-MAC it is
carried out by periodic channel sampling. Normally, a
node is in the sleep mode. It wakes up after a predefined
time has elapsed and checks for the network activity. If
activity is detected, the node powers up and stays awake
to receive the incoming packet. After the reception the
node returns the sleep mode. If no packet is received after
timeout, the node goes back to sleep as well.
Before sending the data, the transmitter sends a preamble.
To reliably receive the data the preamble should be at
least as long as the interval between two subsequent
checking for channel activity. The length of the preamble
has to give enough time to node to wake up, detect the
activity in the channel, and receive the preamble. After
receiving the preamble, the node receives the message. If
the node wakes up and there is no activity at the channel,
the idle listening occurs only for a short duration of time.
The length of the period between two subsequent channel
activity sampling is the result of the trade-off between the
latency and the energy efficiency of the network.
3. DEVELOPMENT OF WIRELESS NODE
The block diagram of the experimental nodes developed
at FME is shown in Fig. 4. It is based on low power
Atmel Atmega16 microcontroller [10] and Microchip
MRF24J40 radio transceiver [11].
Atmega16 [10] is 8-bit microcontroller with 1 KB SRAM
and 16 KB of program memory. It is a low power
microcontroller with power consumption of 1.1mA in the
active mode at (frequency) 1MHz, (voltage) 3V and
(temperature) 25C, and 0.35mA in the idle mode.
Furthermore, the microcontroller has 8 single ended
channels for 10-bit ADC, 32 bidirectional DIO, one 16-bit
and two 8-bit timer/counters with compare modes. The
oscillator frequency is up to 16MHz. In our platform
8MHz has been used.
MRF24J40 [11] transceiver is compliant with
IEEE802.15.4 standard. Its operating frequency band is
2.4GHz and the data rate is 250 Kb/s. It has integrated
20MHz oscillator. Its typical power consumption is 19mA
in receive (Rx), 23mA in transmit (Tx) and 2A in the
sleep mode. Thus, the transceiver’s power consumption in
active modes (Tx or Rx) is around 20 times higher than
the power consumption of the microcontroller.
MCF24J40 operates within Mikroelektronika ZigBee3
development board.
Fig. 4. FME WSN node
The communication between the microcontroller and the
transceiver is carried out using 4-wire SPI interface. SPI
functions in full duplex master/slave configuration. The
microcontroller (master) sends data to the transceiver
trough MOSI, and the transceiver (slave) sends data to the
microcontroller on MISO line. Serial clock (SCLK) is
output from the master. SPI is used not only for data
transfers, but also for transceiver programming using
protocol defined by the manufacturer.
Besides SPI lines, there are three additional lines between
the microcontroller and the transceiver. These are
RESET, WAKE and INT. RESET and WAKE are used
for transceiver reset and wake. Using INT line MRF24J40
transceiver can signal one of 8 available interrupts to the
microcontroller. The most interesting interrupts for our
application are receive interrupt and transmit interrupts
for normal transmission and guarantied time slots
transmissions in a beacon mode.
4. EXPERIMENTAL VALIDATION
The described wireless node, developed at FME, has been
implemented for acceleration monitoring in turning. Fig.
5 shows the experimental setup. The accelerometer is
mounted on a tool holder and it is wired to the wireless
node that is placed on toolpost. The other node is
connected to the PC via DIO. Since the sink (PC) is mains
powered and there is no need for multi-hop configuration,
the non-beacon mode of operation has been chosen.
Although not used in this setup, B-MAC protocol was
implemented to allow extensions to multi-hop
configurations (i.e., topologies) in industrial environments
where physical constraints prohibit use of a single-hop
communication between the sensor and the gateway.
ADXL311 Analog Devices dual-axes measurement
system, with measuring range 2g was used. It is powered
from microcontroller by 5V. In our setup, only vertical
acceleration component is measured. Accelerometer
signal is linked to microcontroller port A0.
Microcontroller performs 10-bit ADC. In the given
installation the maximum sampling rate is 15kHz.
Acceleration measurements have been widely used for
tool condition monitoring and a number of methods for
tool breakage and tool wear have been proposed [1].
These methods were usually based on Discrete Wavelet
Transform, Hilbert-Huang Transform and other
sophisticated techniques. Nevertheless, in this paper the
WSN are in focus, rather than diagnosis techniques. This
is the reason that it was decided to simply employ in-node
signal tresholding.
Fig. 5. Experimental setup
To provide a significant change of acceleration on the tool
holder, we use intermittent turning of grooved part (Fig
5). On the cutting process stop at surface A the high
vibrations occur. In the given setup the acceleration signal
below the value that corresponds to 2.71V is considered
as the event which has to be detected. On this event the
microcontroller wakes up the transceiver and sends a
message containing the signal value (2 bytes) and the
number of occurrences (1 byte). In order to avoid the
detection of transients after the cutting process stops, we
introduce a waiting period of 50ms after an event is
detected, before we allow the detection of the next event.
At least 50ms have to pass between two subsequent
events regardless of the value of the acceleration signal.
On data reception, the sink node puts the port D high and
keeps it high for 20ms. To verify the performance of
nodes, the voltage level of port D and acceleration signal
are acquired using DAQ with 500Hz sampling rate. In this
setup, the DAQ is used as a kind of digital oscilloscope.
The sent data package contains only 3 bytes of data, while
the whole frame for sending the data package contains 22
bytes (Fig. 3). The short addresses that give the
addressing field of 8 bytes are used. The data transmission
assumes SPI communication (2Mbps) of these bytes
between microcontroller and transceiver on both (Rx and
Tx) sides, as well as radio communication (250 Kb/s).
Besides, the microcontroller and the transceiver exchange
control sequences via SPI. This leads to latency of event
detection of 4ms on the receiver’s side.
Fig. 6. Experimental results
Fig. 6 shows the accelerometer signal (acquired by DAQ)
along with the voltage level on the port D of the receiver.
A number of experiments were carried out on Hasse &
Werde lathe. In the experiment shown in Fig. 6 the cutting
parameters were: n=300rpm, s=0.05mm/r and a=0.7mm.
The experimental results indicate a reliable performance
of the developed wireless node.
5. CONCLUSION
In this paper, the application of WSN in machining
operations monitoring has been investigated. Initially, the
communication reliability was questioned. The concern
was that a huge amount of ferrous materials in the
communication area could lead to electro-magnetic
disturbances, causing communication failures and packet
drops. Nevertheless, the experiments have shown that the
communication was reliable. As the distance between the
nodes (connected to the sensor and to the PC) was 2m,
there was no need for multi-hop network deployment.
Using WSNs the placement of sensors at otherwise
inaccessible/difficulty accessible locations such as tool or
even workpiece becomes easily feasible. This opens up
the new possibility in machining process monitoring and
control.
Acknowledgement: The authors express the gratitude
to the Serbian Ministry of Education and Science,
research grants TR35007, TR35020, TR35045, TR35030.
REFERENCES
[1] TETI, R., JEMIELNIAK, K., O’DONNELL, G.,
DORNFELD, D. (2010) Advanced monitoring of
machining operations, CIRP Annals – Manufacturing
Technology, Vol. 59, pp 717-739
[2] TIWARI, A., LEWIS, F., L., GE, S., S. (2008)
Wireless Sensor Network for Machine Condition
Based Maintenance, 8th Int. Conference on Control,
Automation, Robotics and Vision,,pp 461-467
[3] SUPROCK, C., A., NICHOLS J., S. (2009) A Low
Cost Wireless High Bandwidth Transmitter for
Sensor-Integrated Metal Cutting Tools and Process
Monitoring, International Journal of Mechatronics
and Manuf. Systems, Vol. 2, No. 4, pp 441-454
[4] WRIGHT, P., DORNFELD, D., OTA, N. (2008)
Condition Monitoring in End-Milling using Wireless
Sensor Networks
(WSNs), Transactions of
NAMRI/SME, Vol. 36, pp 177-183
[5] PRICKETT, P., SIDDIQUI, R., A., GROSVENOR,
R., I. (2011) A microcontroller-based end milling
cutter monitoring and management system,
International Journal of Advanced Manufacturing
Technology, doi 10.1007/s00170-010-3135-z
[6] SIDDIQUI, R., A., AMER, W., AHSAN, Q.,
GROSVENOR, R., I., PRICKETT, P. (2007) Multiband infinite impulse response filtering using
microcontrollers for e-monitoring applications,
Microprocess and Microsystems, Vol. 31, pp 370-380
[7] IEEE 802.15.4 Standard
[8] DEMIRKOL, I., ERSOY, C., ALAGROEZ, F. (2006)
MAC Protocols for Wireless Sensor Networks: a
Survey, IEEE Communications Magazine, Vol. 44,
No. 4, pp 115-121
[9] POLASTRE, J., HILL, J., CULLER, D. (2004)
Versatile Low Power Media Access for Wireless
Sensor networks, SenSys '04 Proceedings of the 2nd
international conference on Embedded networked
sensor systems, pp 95-107
[10] Atmel (2010) Atmega16 Datasheet
[11] Microchip (2008) MRF24J40 Datasheet